The key insight: most people use AI to answer questions. He used AI to extract structure.
Why Structure Beats Content
Content — facts, definitions, explanations — is easy to find. Wikipedia, textbooks, papers, all accessible.
Structure is harder: What do all experts agree on? Where do they disagree? What's counterintuitive? What's load-bearing for decisions?
Structure is what an expert has after years in a field. It's what lets them navigate new information quickly and know what matters. AI can extract structure from a large corpus of material in ways that would take humans much longer.
The Setup
Load everything you can find into NotebookLM:
- Core textbooks (6 for a serious field)
- Key research papers (15-20)
- Course syllabi and lecture notes
- Review articles and survey papers
More source material = more reliable extraction. Don't skimp.
The Questions That Actually Work
Most people ask AI to "summarize" or "explain." Wrong frame. Ask for structure:
"What are the 5 core mental models that all experts in this field agree on?"
Extracts the settled foundations — what experts take for granted.
"What are the 3 main debates where experts disagree? What does each side believe and why?"
Surfaces the live frontiers and the assumptions that are contested.
"What's the most counterintuitive thing about this field that surprises newcomers?"
Identifies where intuition fails — often the most important things to know.
"If I had to make a real decision in this field tomorrow, what would I most need to know?"
Forces practical prioritization over academic completeness.
The Second Pass
After the initial extraction: ask the AI to generate questions that would reveal whether you actually understand the material. Answer them. Have the AI assess your answers and identify gaps.
This isn't a shortcut to expertise. It's a shortcut to orientation — arriving at week 1 of expertise instead of week 8 of confusion. The depth comes from subsequent work. But starting oriented instead of lost is a significant acceleration.